AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Semantics-Consistent Feature Search for Self-Supervised Visual Representation Learning

Kaiyou Song, Shan Zhang, Zimeng Luo, Tong Wang, Jin Xie

ICCV 2023 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add semantics-consistent feature search (local data to local feature contrast) (ablation)DINO ResNet50 · ImageNet100 linear probinglinear probing top-1: 87.0 -> 89.2Improved
add multi-layer contrast to DINO instead of feature search (ablation)DINO · ImageNet100 linear probinglinear probing top-1: 87.0 -> 87.4Improved
use soft feature search instead of hard ROIAlign crop (ablation)SCFS · ImageNet100 linear probinglinear probing top-1: 88.1 -> 89.2Improved
add feature search to setting without multi-crop (ablation)DINO without multi-crop · ImageNet100 linear probinglinear probing top-1: 83.7 -> 86.3Improved
apply feature search on Res2+Res3 instead of Res2 only (ablation)SCFS · ImageNet100 linear probinglinear probing top-1: 86.1 -> 87.5Improved
apply feature search on Res2+Res3+Res4 instead of Res2+Res3 (ablation)SCFS · ImageNet100 linear probinglinear probing top-1: 87.5 -> 89.2Improved
add SCFS to DINO on ResNet101DINO R101 · ImageNet100 linear probinglinear probing top-1: 86.3 -> 88.3Improved
add SCFS to DINO on ViT-SDINO ViT-S · ImageNet100 linear probinglinear probing top-1: 80.4 -> 81.0Improved
add SCFS to DINO when pre-training on COCODINO ResNet50 · COCO detection after COCO pre-trainingAP^b: 39.0 -> 40.9Improved
add SCFS to DINO (256 bs, 200 epochs)DINO ResNet50 · ImageNet linear probingtop-1 accuracy: 73.0 -> 73.9Improved
add SCFS to DINO for COCO 2x schedule detectionDINO ResNet50 · COCO Mask R-CNN 2xAP^b: 41.9 -> 42.1 (HSA 42.2)Mixed
add SCFS feature searchDINO ResNet50 training cost · training time bs 256, 200 epochstraining time: 147h -> 192hHurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy